Re: Is UML an option for the IETF?

[Tom Scott <tscott@vedatel.com>]

Walter, Avri and others:

1. A combination of text and UML graphics would be a practical
solution, keeping in mind that the "official" version would be the
text and possibly a few simplified "ascii art" representations.
There'ss ample opportunity to refer to outside sources in the primary
documents.

2. As a demo of combined text and graphics, perhaps we could list a
few classes rooted at or near the root, and descend the tree just a
few levels. Something like:

* services (management and end-user oriented)

* sessions, calls, and other temporal activations/groupings of the
services, and the network resources that support them

* hosts, routers, switches, transmission systems, and structured
cabling elements

* tables and other information resources in intermediate systems,
including mappings of interfaces and addresses ("atomic statements"),
and protocols ("non-atomic formulas" and "rules of inference") -- this
analysis could go one step further into a logical model of
routers/switches as first-order theories, but a straightforward
description of the functions and functional groupings would do the job
at this point in time.

* protocols in general

* the computing/communications environment for defining, storing and
executing the obejcts

3. The last category, protocol, would be an appropriate topic, seeing
as how one of the defining characteristics of the Internet is that it
is protocol driven. what is a protocol? Any object or class, not only
protocols but other Internet objects, can be modeled on three levels,
although we should probably work on the surface (leave the deeper
levels to theoreticians for correctness and completeness proofs):

* practical, engineering analysis of protocol: a protocol consists of
functions (or functionality), formatted fields and messages (PDUs),
and finite-state machines that describe the behavior of the protocol
objects.

* mathematical, logical analysis of protcool: ala Mohamed Gouda's
_Elements of Network Protocol Design" (New York: Wiley, 1998), where a
protocol is defined as a network of processes, and processes are
defined as constants, inputs, variables and actions, and (I presume) a
network would be defined as a graph whose nodes are processes and arcs
are channels (ch.p.q == the channel from p to q).

* structural or theoretical analysis of protocol object: don't go
there (yet) ...

Specifically, how would you draw the classification tree for
protocols? What would be the PDUs and their fields, and the FSMs? In
regards to the functions, suppose for the sake of the discussion, that
you take the list of functions from section 10.3 of Stefan Boecking's
_Object-oriented Network Protocols_ (New York: Addison-Wesley, 2000): 

- alignment
- blocking
- byte ordering
-checksumming
- concatenation
- connection control
- encapsulation
- encryption
- error detection/control
- forward error correction
- inactivity contrl
- jitter compensation,
- multiplexing
- (rate) flow control
- relaying
- routing
- segmentation
- signature
- splitting
- traffic control
- traffic padding
- traffic shaping,
- window flow control

Feel free to modify the list or use one that is more appropriate to
this context.

--  Cheers, TT
---------------------------------------------------
Tom Nelson Scott             Vedatel Co
1411 Sheffield Dr.           Bowling Green OH 43402
"In IP We Trust"   "Java Rules"   "E Pluribus Unix"
---------------------------------------------------


-------- Original Message --------
Subject: RE: Is UML an option for the IETF?
Date: Fri, 1 Sep 2000 11:10:14 -0400
From: "Weiss, Walter" <wweiss@ellacoya.com>
To: "'Avri Doria'" <avri@nortelnetworks.com>, NIM List <nim@psg.com>




Avri,

I too have found UML extremely useful for visually representing a
model. While I find it an excellent tool to build concensus 
around the basic organization of objects, I believe we still need a
text based representation to define the actual standard for the 
model. The primary reason is not IETF policy but rather my personal
desire to see tools that take a model as input and a MIB, PIB, 
or whatever as output.

I am not aware of any ASCII based UML tool. Nor do I believe that a
UML in Internet Drafts is terribly useful (primarily because 
of the 72 column limit. However, what I have done in the past is use a
UML package and generate a PDF. The PDF can be posted on a 
server for people to access through their browser or download, print,
and paste.

Since text based models are very difficult to read and conceptuallize,
I would prefer to have something like this as backup material 
that is referenced by a draft but is not specifically part of a
standard.

regards,

-Walter

> -----Original Message-----
> From: Avri Doria [mailto:avri@nortelnetworks.com]
> Sent: Friday, September 01, 2000 9:17 AM
> To: NIM List
> Subject: Is UML an option for the IETF?
> 
> 
> Hi,
> 
> In regard to Tom's suggestion that UML be used to  describe the 
> model:
> 
> Is it possible to use UML  without graphics support?  And if not,
> can we justifiably use it as the IETF method of description. 
> Personally, I am getting comfortable  with UML when using UML drawing 
> packages (e.g. Rose).  But I find it  very uncomfortable to use with 
> ascii characters.  As far as I can tell UML is totally implementation
> dependent; i.e. if you can't draw it, you  can't express it.
> 
> Is there a way around this problem?  Other by writing a graphic UML 
> to ascii character drawing converter for each UML graphics program?
> 
> a.
> -- 
> 
> Avri Doria
> +1 401 663 5024
>